For Economistss

20 Practical Ideas for Economists to Stay Cognitively Sovereign

AI models can generate plausible-looking economic specifications with assumptions you have not tested. Weak scrutiny of these assumptions leads to policy recommendations built on hidden data fitting rather than theory.

These are suggestions. Take what fits, leave the rest.

Download printable PDF

Testing Model Assumptions

Document every assumption before running AIbeginner

Write your causal story first. Then check what AI suggests against it.

Ask AI to state its functional form choicebeginner

Demand explicit explanation of why linear, log-linear, or nonlinear specification.

Reverse-engineer lag structures AI selectsintermediate

Why did it choose 12 months not 6. Economic theory or data fit.

Test coefficient signs against economic logicbeginner

Before celebrating R-squared, check elasticities make sense theoretically.

Require AI to name excluded variablesintermediate

Ask what it left out and why. Omitted variable bias is structural.

Run AI models on random data subsetsintermediate

Does the specification hold across time periods or just your chosen window.

Compare AI output to published peer estimatesbeginner

Your coefficient should fall within the plausible range from literature.

Ask for parameter stability across subsamplesintermediate

Do results change when you drop sectors, regions, or demographic groups.

Identify which assumptions drive policy conclusionsintermediate

Sensitivity analysis on elasticities, not just statistical significance.

Reject models with unexplained fit improvementsundefined

If R-squared jumps 20 percent, understand why before citing it.

Defending Forecasting Judgement

Set confidence intervals before AI generates themintermediate

Your prior bounds on uncertainty. Compare to what model claims.

Demand conditional forecasts not just point estimatesintermediate

Show outcomes under recession, stagflation, baseline scenarios separately.

Name the structural breaks AI model ignoresintermediate

Policy regime changes, technological shifts, demographic breaks affect future.

Track what AI predicted wrong last quarterbeginner

Keep a record of its forecast errors. Recognize the pattern.

Require AI to state its training data windowbeginner

Models trained only on 2010 to 2023 will miss structural shifts.

Separate model output from your judgement callintermediate

AI predicts 2.1 percent growth. You decide if 1.8 to 2.4 is reasonable.

Test forecasts against leading indicators you trustintermediate

Does AI projection align with yield curves, PMI, labour flows.

Build your own simple benchmark model annuallyintermediate

Naive forecasts force you to explain where AI models add real value.

Present forecast scenarios to non-economist stakeholdersundefined

Their pushback reveals hidden assumptions you normalised.

Document why you rejected AI forecast recommendationsbeginner

Write this down. Accountability sharpens your independent economic reasoning.

Five things worth remembering

AI is fastest at generating plausible-sounding prose around weak assumptions.
Confidence intervals from AI models are statistical, not epistemic.
Your theoretical training is your only defence against data fitting.
Economic models that cannot be explained in plain words hide problems.
Forecasts from AI gain authority through presentation, not accuracy.

20 Practical Ideas for Economists to Stay Cognitively Sovereign

Testing Model Assumptions

Defending Forecasting Judgement

Cognitive Sovereignty: How To Think For Yourself When AI Thinks For You